Secure, Automated, Internet-Based mmWave Test and Measurement with Xilinx RFSoC
Luc F. Langlois, Avnet
Fabrício Dourado, Rohde & Schwarz
Wireless system design at mmWave frequencies presents unique challenges, calling for multidisciplinary engineering teams scattered across geography and home offices with deep expertise in RF system design, digital signal processing, embedded software, and emerging wireless standards.
Join engineers from Avnet and Rohde & Schwarz as they demonstrate test solutions for Avnet’s Wideband mmWave Radio Development Kit for Xilinx Zynq® UltraScale+™ RFSoC Gen-3. You will observe automated measurement techniques based in MATLAB®using Rohde & Schwarz’s 5G NR signal generation and analysis for fast, remote testing through the new Rohde & Schwarz Secure Application Gateway™.
Explore the tradeoffs of testing critical wireless metrics, including ACLR and EVM, both 5G NR and standard-agnostic, from R&D to production. We will also investigate the performance of the bits-to-mmWave radio in terms of frequency response and distortion for frequency planning or optimizing the system design.
Published: 26 May 2022
Welcome to Secure, Automated, Internet-based Millimeter Wave Testing with Xilinx RFSoC from MATLAB Expo. I'm Luc Langlois from Avnet and I'm here with my friend and collaborator, Fabricio Dorado from Rohde & Schwarz. Technology design cycles, they're expanding. We're seeing two plus years for Xilinx RFSoC designs with wireless systems. They tend to proceed in two phases.
The first phase is a proof of concept, and that is typically initiated by the system architect long before the product definition. System architects have a wish list. They want system level abstraction modeling. They don't want to mobilize FPGA and software teams for any low level programming, and they want some sort of automated control to verify their concepts quickly from the hardware on test equipment.
And these are the attributes of our system under test. Decisions from the system architect tend to influence the design cycle from the very beginning all the way through system level modeling, simulation, FPGA design, software, and even test and validation. At the same time, multi-disciplinary engineering design teams are now distributed globally. Some may even be working at home.
Given the cost and complexity of millimeter wave technology, it's unrealistic to replicate the system under test for each member of a large team. Ideally, they want to have access to the centralized hardware systems and test equipment for collaborative development. Of course, this requires secure links when uploading sensitive information, source code, and even proprietary waveforms to the system under test. This is where security is such a big concern.
So here's our solution. Secure automated internet based testing with Xilinx RFSoC. It starts with Avnet RFSoC Explorer, a MATLAB based app for streaming standard compliant waveforms such as 5G and any other custom waveforms to the board, which is the Avnet Wideband millimeter Wave Radio Development Kit for Xilinx RFSoC.
In conjunction, Rohde & Schwarz provides MATLAB support for test instrument control for seamless workflows from algorithm to deployment with precise, repeatable measurements from the system under test at every stage of the development.
The system under test is hosted in the Rohde & Schwarz Secure Application Gateway, providing secure, uninterrupted access to distributed engineering teams throughout the world. All the data links are encrypted and access protected by the LANCOM Rohde & Schwarz UTM Firewall for cybersecurity levels that are really equivalent to those of modern financial transactions.
The system under test is controlled over the LAN through RFSoC Explorer, giving you the ability to natively perform analysis, create simulation models that interact with streaming data while remaining in the MATLAB environment. The system under test is the Avnet millimeter Wave Development Kit for Xilinx ZYNQ RFSoC Gen-3, and that's the evaluation board, the ZCU208 with eight channels of direct sampling gigasample per second ADCs and DACs coupled to the Otava DTRX2 up and down converter to and from millimeter wave.
Traditional test and measurement may have involved various techniques such as looping back the transmit to the receiver. That works for some 6 gigahertz applications if you don't have any test equipment. Or, of course, just sending your transmit signal over the air from the board to the test equipment. But really, that was lacking any closed loop instrument control or MATLAB integration. So what we've done in the next slide is put together this system that is hosted in the Rohde & Schwartz Secure Application Gateway remotely.
And the system is driven, as I mentioned, from the MATLAB application, which is Avnet's RFSoC Explorer. And this allows working with MATLAB-based waveforms such as the MathWorks 5G Toolbox that can send standard compliant 5G NR waveforms through the RFSoC DACs mixed up to millimeter wave between 19 and 31 gigahertz, and to the FSW spectrum analyzer for various personality applications to run some analysis and to display the spectrum to demodulate and such, as Fabricio will show you very shortly in the demo. Over to you, Fabricio.
Yes. Thank you, Luc. Great. Yeah, so we see here also in this slide, we are going to demonstrate for you the testing of the test signal chain, but we already prepared with the six generator part to test also RX. And eventually in the future, we will also integrate our Signal Explorer, which would allow you to test the receiver pass exactly with the same features that we are testing with the FSW.
And here you see our test setup. I mean, this is the setup that Luke and I, and also you see it from Otava, are using to develop these to integrate our equipment into a MATLAB-based app from Avnet. And because we want fully access externally and would like to allow the developers to be able to change connections remotely, we performed here very good calibration of the total system, including also our millimeter wave switch.
So in this next slide you see here. So basically, you have the board connected to all ports, and also the analyzing signal generator are connected to each other. We have even some port here for training purposes to show people how can you use the user correction from the signal generator to perform fast alignment without having to write software. So this works really fine, and I tested-- I verified the performance of the setup in terms of 5G NR EVM with and without the switch, and we are getting exactly the same performance.
So this high end frequency response correction that we have for the SMW and FSW is really useful here. And here you see our challenge. Our challenge here is that we have both to characterize the device in the digital and the analog domain. And typically if you want to characterize, for example, a frequency response channel response in magnitude and phase, you typically do it with VNA. But how to connect the VNA to the digital domain, right?
So we are doing this here with an option called FSW K18. It's actually amplifier testing personality, but with that one we can also test duties like this. So digital to our RF, in this case, digital to millimeter wave DUTs. And here is our setup for the demo. So I'm going to present two type of measurements for you, the EVM, SLR, and spectral emission mask combined in a single measurement.
This is interesting not only for DUTs, for Devices Under Test, without 3 [INAUDIBLE] capability, but also, for example, in OTA cases and then the channel response with the K18. So let's take a look on this. Typically when you test, according to a GPP, your capture time is 20 milliseconds. And if you test all these slides, these tests will take very long. And you see here we have a good standard deviation for the EVM, results and this is because we are always triggering exactly at the same frame start of the waveform.
And this allows us, for example, to test less slots per frame and to perform those measurements faster. I compared testing all slides with testing, for example, only one slot, and you increase the standard deviation. I'll show you that in a minute. But the point is in the R&D sometimes you want to do a little bit faster measurements. You'll see now I would change to one slot and a much a shorter capture time-- now 3.5 milliseconds.
And you are going to see that the measurement results will be much faster. And I think considering that we are not testing all these slides-- all these slots, the standard deviation that we get for EVM is still really, really good even testing only one slot. Becomes around 0.2%, 0.3%, but with the advantage that you test all this in a fraction of the time. So just for comparison purposes, I put you here the original methods you measure.
And let's say I'm measuring here now 300 microseconds and you see that it's even much shorter than we had before, 10 times-- more than 10 times shorter than the previous one, and then you get this kind of variation. But this allows you, in the R&D, to make much faster power sweeps, frequency sweeps, and analyze. And then later on, if you are at the end of your process, then you can go for compliance testing strictly according to [INAUDIBLE].
Now, let's take a look in the channel frequency response. You see here we have the K18. I'm measuring here. The measurements are running, and I'm getting very similar EVM, although the method here to measure EVM is totally different. And I just turned off the channel-- the equalization of the channel.
So you see here we are able, in magnitude and phase, to fully characterize digital to our devices, and we are doing that because we are loading the waveform as a reference signal here. So quite useful, and this is kind of a type of characterization typically not available with signal analyzer.
And let's say you like this equalization here. You want to use it. Maybe in the future somebody will ask you, ask us, well, we have so much capability with RFSoC and the tools from MathWorks. Can we get this proficient? Yes, we can.
So you go there to save equalizer, and then you're going to see that in example, and you can save it in the [INAUDIBLE] format or as a text file. And then you can take this and maybe in the future, somebody asks us to implement this here. And for that one, Luc, I think you are better in a position to explain us what you mean with synthesizable HDL by Avnet.
Yes, thank you, Fabricio. This is a potential application that one could implement with the system that we have here. Fabricio just mentioned the channel equalization that is done in the K18 personality module. That can be exported and we could, for example, take the channel impulse response and implement a pre-equalizer using some sort of adaptive finite impulse response filter quite easily in the RFSoC programmable logic.
And the way that we would do that is with MathWorks HDL Coder, for which we have a support package for the system. So this is something, an example, that we could implement and it would be quite useful to have the pre-equalizer directly in the RFSoC device on the board. So finally, we invite you to learn more. We are hosting some demos and training and you can register for more information at the link below.
So in conclusion, we have shown you millimeter wave systems requiring secure, robust test automation, our application, the Avnet RFSoC Explorer for enabling the AMD Xilinx RFSoC within MATLAB with Rohde & Schwarz MATLAB support for instrument in the loop standard compliant testing, and finally, the Rohde & Schwarz Secure Application Gateway, enabling worldwide collaborative remote access for distributed engineering teams in real time with state of the art data security. So last slide. Thank you very much for attending MATLAB Expo I think there's a last one, Fabricio.
Thank you.
And hope to have some further discussions with you in our respective booths. Please feel free to chat with us. We are standing by to host any questions right now.